When an LLM writes an essay about, say, Oppenheimer, the essay it spits out isn't, word for word, in the training data. it's created this new essay. Following that, asking it to write an essay about Doofenhimer, a fictional contemporary of Oppenheimer that I just invented, where was it trained on for that? to say it can't create anything when, given the depth of the training data, it seems capable of creating things, seems like an underestimation of its abilities, and moreover, what it actually means to be creative. when I'm being creative, there are connections being made between completely unrelated things that I've been exposed to at some point in my past which I combine in creative ways to create new things. if it's all just pattern recognition, interpolation, and recombination, then either nothing new can ever be created by humans doing exactly that, or LLMs are also creating new things. Are Salvador Dali's paintings not creative because the idea of clocks and the concept of melting aren't new?
https://chat.openai.com/share/2b9a8245-2063-4722-81e8-20d10c...